xref: /openbmc/linux/net/ipv6/inet6_hashtables.c (revision e23feb16)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Generic INET6 transport hashtables
7  *
8  * Authors:	Lotsa people, from code originally in tcp, generalised here
9  * 		by Arnaldo Carvalho de Melo <acme@mandriva.com>
10  *
11  *	This program is free software; you can redistribute it and/or
12  *      modify it under the terms of the GNU General Public License
13  *      as published by the Free Software Foundation; either version
14  *      2 of the License, or (at your option) any later version.
15  */
16 
17 #include <linux/module.h>
18 #include <linux/random.h>
19 
20 #include <net/inet_connection_sock.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet6_hashtables.h>
23 #include <net/secure_seq.h>
24 #include <net/ip.h>
25 
26 int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
27 {
28 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
29 	int twrefcnt = 0;
30 
31 	WARN_ON(!sk_unhashed(sk));
32 
33 	if (sk->sk_state == TCP_LISTEN) {
34 		struct inet_listen_hashbucket *ilb;
35 
36 		ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
37 		spin_lock(&ilb->lock);
38 		__sk_nulls_add_node_rcu(sk, &ilb->head);
39 		spin_unlock(&ilb->lock);
40 	} else {
41 		unsigned int hash;
42 		struct hlist_nulls_head *list;
43 		spinlock_t *lock;
44 
45 		sk->sk_hash = hash = inet6_sk_ehashfn(sk);
46 		list = &inet_ehash_bucket(hashinfo, hash)->chain;
47 		lock = inet_ehash_lockp(hashinfo, hash);
48 		spin_lock(lock);
49 		__sk_nulls_add_node_rcu(sk, list);
50 		if (tw) {
51 			WARN_ON(sk->sk_hash != tw->tw_hash);
52 			twrefcnt = inet_twsk_unhash(tw);
53 		}
54 		spin_unlock(lock);
55 	}
56 
57 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
58 	return twrefcnt;
59 }
60 EXPORT_SYMBOL(__inet6_hash);
61 
62 /*
63  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
64  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
65  *
66  * The sockhash lock must be held as a reader here.
67  */
68 struct sock *__inet6_lookup_established(struct net *net,
69 					struct inet_hashinfo *hashinfo,
70 					   const struct in6_addr *saddr,
71 					   const __be16 sport,
72 					   const struct in6_addr *daddr,
73 					   const u16 hnum,
74 					   const int dif)
75 {
76 	struct sock *sk;
77 	const struct hlist_nulls_node *node;
78 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
79 	/* Optimize here for direct hit, only listening connections can
80 	 * have wildcards anyways.
81 	 */
82 	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
83 	unsigned int slot = hash & hashinfo->ehash_mask;
84 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
85 
86 
87 	rcu_read_lock();
88 begin:
89 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
90 		if (sk->sk_hash != hash)
91 			continue;
92 		if (likely(INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
93 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
94 				goto begintw;
95 			if (unlikely(!INET6_MATCH(sk, net, saddr, daddr,
96 						  ports, dif))) {
97 				sock_put(sk);
98 				goto begin;
99 			}
100 		goto out;
101 		}
102 	}
103 	if (get_nulls_value(node) != slot)
104 		goto begin;
105 
106 begintw:
107 	/* Must check for a TIME_WAIT'er before going to listener hash. */
108 	sk_nulls_for_each_rcu(sk, node, &head->twchain) {
109 		if (sk->sk_hash != hash)
110 			continue;
111 		if (likely(INET6_TW_MATCH(sk, net, saddr, daddr,
112 					  ports, dif))) {
113 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
114 				sk = NULL;
115 				goto out;
116 			}
117 			if (unlikely(!INET6_TW_MATCH(sk, net, saddr, daddr,
118 						     ports, dif))) {
119 				sock_put(sk);
120 				goto begintw;
121 			}
122 			goto out;
123 		}
124 	}
125 	if (get_nulls_value(node) != slot)
126 		goto begintw;
127 	sk = NULL;
128 out:
129 	rcu_read_unlock();
130 	return sk;
131 }
132 EXPORT_SYMBOL(__inet6_lookup_established);
133 
134 static inline int compute_score(struct sock *sk, struct net *net,
135 				const unsigned short hnum,
136 				const struct in6_addr *daddr,
137 				const int dif)
138 {
139 	int score = -1;
140 
141 	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
142 	    sk->sk_family == PF_INET6) {
143 		const struct ipv6_pinfo *np = inet6_sk(sk);
144 
145 		score = 1;
146 		if (!ipv6_addr_any(&np->rcv_saddr)) {
147 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
148 				return -1;
149 			score++;
150 		}
151 		if (sk->sk_bound_dev_if) {
152 			if (sk->sk_bound_dev_if != dif)
153 				return -1;
154 			score++;
155 		}
156 	}
157 	return score;
158 }
159 
160 struct sock *inet6_lookup_listener(struct net *net,
161 		struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
162 		const __be16 sport, const struct in6_addr *daddr,
163 		const unsigned short hnum, const int dif)
164 {
165 	struct sock *sk;
166 	const struct hlist_nulls_node *node;
167 	struct sock *result;
168 	int score, hiscore, matches = 0, reuseport = 0;
169 	u32 phash = 0;
170 	unsigned int hash = inet_lhashfn(net, hnum);
171 	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
172 
173 	rcu_read_lock();
174 begin:
175 	result = NULL;
176 	hiscore = 0;
177 	sk_nulls_for_each(sk, node, &ilb->head) {
178 		score = compute_score(sk, net, hnum, daddr, dif);
179 		if (score > hiscore) {
180 			hiscore = score;
181 			result = sk;
182 			reuseport = sk->sk_reuseport;
183 			if (reuseport) {
184 				phash = inet6_ehashfn(net, daddr, hnum,
185 						      saddr, sport);
186 				matches = 1;
187 			}
188 		} else if (score == hiscore && reuseport) {
189 			matches++;
190 			if (((u64)phash * matches) >> 32 == 0)
191 				result = sk;
192 			phash = next_pseudo_random32(phash);
193 		}
194 	}
195 	/*
196 	 * if the nulls value we got at the end of this lookup is
197 	 * not the expected one, we must restart lookup.
198 	 * We probably met an item that was moved to another chain.
199 	 */
200 	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
201 		goto begin;
202 	if (result) {
203 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
204 			result = NULL;
205 		else if (unlikely(compute_score(result, net, hnum, daddr,
206 				  dif) < hiscore)) {
207 			sock_put(result);
208 			goto begin;
209 		}
210 	}
211 	rcu_read_unlock();
212 	return result;
213 }
214 
215 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
216 
217 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
218 			  const struct in6_addr *saddr, const __be16 sport,
219 			  const struct in6_addr *daddr, const __be16 dport,
220 			  const int dif)
221 {
222 	struct sock *sk;
223 
224 	local_bh_disable();
225 	sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
226 	local_bh_enable();
227 
228 	return sk;
229 }
230 
231 EXPORT_SYMBOL_GPL(inet6_lookup);
232 
233 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
234 				     struct sock *sk, const __u16 lport,
235 				     struct inet_timewait_sock **twp)
236 {
237 	struct inet_hashinfo *hinfo = death_row->hashinfo;
238 	struct inet_sock *inet = inet_sk(sk);
239 	const struct ipv6_pinfo *np = inet6_sk(sk);
240 	const struct in6_addr *daddr = &np->rcv_saddr;
241 	const struct in6_addr *saddr = &np->daddr;
242 	const int dif = sk->sk_bound_dev_if;
243 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
244 	struct net *net = sock_net(sk);
245 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
246 						inet->inet_dport);
247 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
248 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
249 	struct sock *sk2;
250 	const struct hlist_nulls_node *node;
251 	struct inet_timewait_sock *tw;
252 	int twrefcnt = 0;
253 
254 	spin_lock(lock);
255 
256 	/* Check TIME-WAIT sockets first. */
257 	sk_nulls_for_each(sk2, node, &head->twchain) {
258 		if (sk2->sk_hash != hash)
259 			continue;
260 
261 		if (likely(INET6_TW_MATCH(sk2, net, saddr, daddr,
262 					  ports, dif))) {
263 			tw = inet_twsk(sk2);
264 			if (twsk_unique(sk, sk2, twp))
265 				goto unique;
266 			else
267 				goto not_unique;
268 		}
269 	}
270 	tw = NULL;
271 
272 	/* And established part... */
273 	sk_nulls_for_each(sk2, node, &head->chain) {
274 		if (sk2->sk_hash != hash)
275 			continue;
276 		if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif)))
277 			goto not_unique;
278 	}
279 
280 unique:
281 	/* Must record num and sport now. Otherwise we will see
282 	 * in hash table socket with a funny identity. */
283 	inet->inet_num = lport;
284 	inet->inet_sport = htons(lport);
285 	sk->sk_hash = hash;
286 	WARN_ON(!sk_unhashed(sk));
287 	__sk_nulls_add_node_rcu(sk, &head->chain);
288 	if (tw) {
289 		twrefcnt = inet_twsk_unhash(tw);
290 		NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
291 	}
292 	spin_unlock(lock);
293 	if (twrefcnt)
294 		inet_twsk_put(tw);
295 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296 
297 	if (twp) {
298 		*twp = tw;
299 	} else if (tw) {
300 		/* Silly. Should hash-dance instead... */
301 		inet_twsk_deschedule(tw, death_row);
302 
303 		inet_twsk_put(tw);
304 	}
305 	return 0;
306 
307 not_unique:
308 	spin_unlock(lock);
309 	return -EADDRNOTAVAIL;
310 }
311 
312 static inline u32 inet6_sk_port_offset(const struct sock *sk)
313 {
314 	const struct inet_sock *inet = inet_sk(sk);
315 	const struct ipv6_pinfo *np = inet6_sk(sk);
316 	return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
317 					  np->daddr.s6_addr32,
318 					  inet->inet_dport);
319 }
320 
321 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
322 		       struct sock *sk)
323 {
324 	return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
325 			__inet6_check_established, __inet6_hash);
326 }
327 
328 EXPORT_SYMBOL_GPL(inet6_hash_connect);
329